Think of a military drone. What comes to mind is probably the classic example from America’s post-9/11 “global war on terror”: the Predator, a fixed-wing behemoth with a 17-metre wingspan and a piston engine. The iconic drone of the first weeks of the war in Ukraine was similarly imposing. The Turkish-made Bayraktar TB2 stretched 12 metres across and flew as high as 25,000 feet. Such machines are scarce in Ukraine these days. The threat from Russian air defences and jamming is so severe that no one wants to risk a large and expensive drone. When one does get into the air, it keeps a distance from the front lines.
Small and cheap drones are a different matter. Our reporting this week examines First Person View (FPV) drones: compact explosive-laden devices, which cost around $400 a pop, flown by nearby pilots who wear goggles and get live video feeds from the flying machines. Data collected by Tochnyi, a collective project, suggest that monthly FPV strikes on infantry have gone from a handful last summer to more than 1,100 in January, with Russia launching just over half of those. These are destroying tanks, artillery pieces, bunkers and troops on both sides. Their importance was underscored this week when Volodymyr Zelensky, Ukraine’s president, ordered the armed forces to create an entirely new branch: the Unmanned Systems Force. I reckon that is the first time a country has introduced a new branch of the armed forces in wartime since Britain created the Royal Air Force in April 1918.
FPV drones are fast, manoeuvrable and difficult to evade. Their low cost means that they can be procured in large numbers—Ukraine wants to build more than 1m this year. Remarkably, that is not far off its annualised current rate of shell consumption. The drones are not unstoppable, though. They have short ranges—tens of kilometres—carry a small payload relative to an artillery shell and can be halted by jamming, which often kicks in as the drone gets within 150 metres of the target.
Both sides are making huge efforts to get more drones through to hit their enemies. Part of that is about pilot skill. Less experienced pilots can manoeuvre their drones to hit targets 10% of the time; good ones have a hit rate more like 70-80%. It is also about making the drones more resistant to jamming. Radios that can change frequencies quickly are key to this, though their details are highly sensitive. Enabling the drone to home in on a target even if the link to the pilot is jammed, however, is the ultimate goal. Both sides are experimenting with the use of artificial intelligence for this kind of autonomous target detection.
On the face of it, this seems trivial. Commercially available drones, like consumer-grade quadcopters, have had rudimentary object detection and tracking capabilities for many years. But people involved in developing this technology in Ukraine tell me it remains a serious unsolved problem—tracking a fast-moving military vehicle, which may be camouflaged and spewing out smoke, is hard, and doing so with a low-cost computer chip that can be bought at scale is even harder.
What’s clear is that the experimentation under way in Ukraine is likely to accelerate the development of cheap and lethal semi-autonomous weapons. And that poses difficult ethical and legal questions. The use of algorithms to allow a munition to lock on to a target in the last phase of flight is hardly new—advanced missiles have done that for decades. But doing so on a far larger scale than ever before, using neural networks that may not have been through extensive testing, is a different matter. And whereas the Ukrainian front lines are relatively static and uncluttered, the same “terminal guidance” could be far riskier to use in areas—like Gaza—where civilians and civilian vehicles are scattered among the combatants.
Thank you for reading Simply Science.